Experimental and clinical evaluation of iterative reconstruction (OSEM) in dynamic PET: Quantitative characteristics and effects on kinetic modeling

Research output: Contribution to journalArticleAcademicpeer-review

165 Citations (Scopus)


The purpose of this study was to investigate the quantitative properties and effects of ordered-subset expectation maximization (OSEM) on kinetic modeling compared with filtered back-projection (FBP) in dynamic PET studies. Both phantom and patient studies were performed. Methods: For phantom studies dynamic two-dimensional emission scans with 10-min frames and 20-min scan intervals were acquired over a 14-h period using an HR+ PET scanner. Various phantoms were scanned: 2-, 5-, 10-, and 20-cm-diameter phantoms filled with an 18F solution (300 kBq/mL) and a NEMA phantom filled with an 18F background (40 kBq/mL) and a cold or 11C insert (450 kBq/mL). Transmission (Tx) scans of 5-60 min were acquired. Data were reconstructed using FBP Hanning 0.5 and OSEM with 2-12 iterations and 12 or 24 subsets. Quantitative accuracy and noise characteristics were assessed. For patient studies, five cardiac, three oncologic, and three brain dynamic 18F-FDG scans were used. Five reconstructions were performed: FBP Hanning 0.5, and OSEM 2 x 12 and OSEM 4 x 16 with and without 5-mm full width at half maximum smoothing. Time-activity curves were calculated using volumes of interest. The input function was derived from arterial sampling. Metabolic rate of glucose (MRglu) was calculated with a standard two-tissue compartment model and Patlak analysis. Results: Contribution of Tx noise to the reconstructed image was smaller for OSEM than for FBP. Differences in signal-to-noise ratio between FBP and OSEM depended on number of iterations and phantom size. Bias with OSEM was observed for regions enclosed within a 5-to 10-fold hotter background. For cardiac studies OSEM 2 x 12 and OSEM 4 x 16 resulted in 13% and 21% higher pixel values and 9% and 15% higher MRglu values compared with FBP. Smoothing decreased all these values to 2%. Similar results were found for most tumor studies. For brain studies MRglu of FBP and OSEM 4 x 16 agreed within 2%. Use of OSEM image-derived input functions for cardiac PET studies resulted in a decrease in calculated MRglu of about 15%. Conclusion: For most PET studies OSEM has equal quantitative accuracy as FBP. The higher pixel and MRglu values are explained by the better resolution of OSEM. However, OSEM does not provide accurate image-derived input functions for FDG cardiac PET studies because of bias in regions located within a hotter background.

Original languageEnglish
Pages (from-to)808-817
Number of pages10
JournalJournal of nuclear medicine
Issue number5
Publication statusPublished - 29 May 2001


  • Dynamic PET
  • Iterative reconstruction
  • Kinetic modeling
  • Quantitative accuracy

Cite this